Quick Start

Introduction to DeepChrInteract

Though deep learning methods have been widely developed for predicting chromatin interactions using flanking DNA sequence in identified chromatin interaction regions, a comprehensive software toolkit to integrate and evaluate different deep learning architectures are under-developed.

The modern project keeps that original motivation and extends it into a PyTorch-based benchmark and research framework for enhancer-promoter interaction prediction.

System requirements

The original documentation listed the following baseline environment:

CPU memory is recommended as 16GB
GPU memory is recommended as 8GB
Python 3.8
Keras == 2.4.0
TensorFlow == 2.3.0
numpy >= 1.15.4
scipy >= 1.2.1
scikit-learn >= 0.20.3
seaborn >=0.9.0
matplotlib >=3.1.0

The current repository uses a different runtime:

Python 3.10+ is recommended;
PyTorch 2.x;
numpy, scikit-learn, matplotlib, tqdm;
transformers for DNA language model backbones;
optional mamba-ssm in CUDA environments.

Installation

Clone the project and install dependencies:

git clone <your-repository-url>
cd Enhancer-Promoter-Interaction
pip install -r requirements.txt

Optional dependency for the Mamba model:

pip install mamba-ssm

Data preprocessing

The current pipeline consumes raw text sequence files and converts them into train.npz, val.npz, and test.npz splits without generating PNG intermediates.

python scripts/preprocess.py \
    --raw_dir data/raw \
    --cell_type GM12878 \
    --out_dir data

Pipeline validation without real data

The repository includes a dummy-mode validation path for testing the training and evaluation pipeline before real biological data are available.

python scripts/test_pipeline.py
python scripts/test_pipeline.py --quick

Single experiment training

python -m src.train \
    --model_id M2 \
    --exp_id E03 \
    --encoding_mode onehot \
    --fusion_strategy concat_sub_mul \
    --cell_type GM12878 \
    --seed 0

Evaluation

python -m src.evaluate \
    --model_id M2 \
    --exp_id E03 \
    --encoding_mode onehot \
    --cell_type GM12878 \
    --seed 0

Five-seed batch experiment

bash scripts/run_experiment.sh E03 M2 GM12878 onehot concat_sub_mul

DNA language model workflow

For M13, embeddings can be precomputed once and reused:

python -c "
from src.encoders import LLMEncoder
enc = LLMEncoder('dnabert2')
# Load enhancer/promoter sequences from processed data and call encode_dataset()
"

MAE pretraining workflow

python -m src.train --model_id M14 --exp_id E16 --pretrain
python -m src.train --model_id M14 --exp_id E16

Documentation deployment

This project is intended to be published as a static documentation site through GitHub Pages after Sphinx builds the HTML output.